Identification of informative subpathways and genes using improved differential expression analysis for pathways method

Abstract

Pathway-based analysis is introduced to define useful biological knowledge by considering the whole pathway features. However, most of these analyses have several shortcomings, such as less sensitivity towards data that could lead to some important information being missed. Because of the deficiency, pathway-based analysis has been shifted to subpathway-based analysis, which is seen to be more relevant in understanding the biological reactions. This is strengthened by the fact that several studies have found abnormalities in pathways caused by certain regions that respond in the etiology of diseases. In addition, subpathway-based analysis has been found to be more effective and sensitive than the whole pathway. Due to this orientation, many tools have been developed to accomplish the inadequate interpretation in biology system. The Differential Expression Analysis for Pathway (DEAP) is one of the methods in subpathway-based analysis which identifies a local region perturbed by complex diseases in large pathway data. However, the method has shown low performance in identifying informative pathway and subpathway. Hence, this research proposes a modified DEAP method (termed iDEAP) for enhancing the identification of perturbed subpathways in pathway activities and aimed at achieving higher performance in the detection of differential expressed pathways. To this end, firstly, asearch algorithm adapted from DMSP algorithm was implemented to DEAP in search for informative subpathways. Secondly, the relation among subpathways was taken into account by averaging the maximum absolute value (termed DEAP score) to emphasize the reaction among subpathways so that efficient identification of informative pathways can be achieved. Three gene expression data sets were applied in this study (head and neck tumour, colorectal cancer and breast cancer). The results were obtained in terms of the number of differential expressed pathways (head and neck tumor-81 pathways, colorectal cancer-78 pathways, breast cancer-95 pathways) and they suggest that the proposed method yielded better performance as compared to previous work. In fact, when the selected genes from the results were evaluated using 10-fold CV in terms of accuracy, the proposed method showed higher accuracy for Colorectal (90%) and Breast cancer (94%). Finally, a biological validation was conducted on the top five (5) significant pathways and selected genes based on biological literatures.